Trained model generation method, trained model, anomaly factor estimation device, board processing device, anomaly factor estimation method, learning method, learning device, and learning data creation method

摘要

PROBLEM TO BE SOLVED: To provide a learned model generation method capable of accurately identifying the cause of an abnormality of a substrate to be processed after processing by a processing fluid and shortening the time for identifying the cause of the abnormality. SOLUTION: A trained model generation method is a machine learning of a trained model LM for estimating a cause of abnormality of a processing target substrate W2 after processing by a processing fluid and a step S31 for acquiring training data TND. Includes step S32, which is generated by doing so. The learning data TND includes the feature amount XD and the abnormality factor information YD. The abnormality factor information YD indicates the cause of the abnormality of the learning target substrate W1 after the processing by the processing fluid. The feature quantity XD is the first feature quantity information XD1 indicating the characteristics of the time transition of the section data SX of the time series data TD1 indicating the physical quantity of the object used by the substrate processing apparatus 200 that processes the learning target substrate W1 by the processing fluid. including. The first feature amount information XD1 is represented by time. [Selection diagram] FIG. 20